Digital Effects I

## Delay * One capability of computers that we haven't talked about much is the ability to store a signal and give back a delayed copy of it * This is pretty much unobtanium for analog systems * Many of the effects we will be looking at make heavy use of this * Data structure is a *queue* of samples: typical implementation is a *ring buffer* ## Ring Buffer Example len = 100 buffer = [0]*len head = 0 tail = 0 empty = True def queue(s): assert empty or head != tail buffer[tail] = s tail = (tail + 1) % len empty = False def dequeue(): assert not empty s = buffer[head] head = (head + 1) % len empty = head == tail return s ## Effect * Term is used lots of ways, but basic idea… * Take in a sound, modify it in a specific way, put it out * Often "realtime" effects, so latency and throughput matter * Realtime effects can be used "offline" too ## LADSPA * "Linux Audio Developers Simple Plugin API" * Example of effects "plugin" architecture: c.f. LV2, VSP, Juce, etc etc * Idea: Provide loadable "modules" with a known API for * Accepting sample streams * Accepting control streams (same thing?) * Accepting control parameters * Emitting output streams * Providing GUI information for rendering * Plugins are available from many places, shipped with Linux distros, etc * Plugin "host" is responsible for loading and plugging together plugins: we will use Audacity for non-realtime demos * Global plugin ID registry ## An Example Plugin: mu-law * Let's look at a plugin and try it out * Effect: [μ-law](Μ-law_algorithm) encoding * Used for telephony to raise quiet parts of signal above line noise * Simple time-domain transformation: no history at all * How does it sound? ## Implementing In SWH * Uses XML as a metadata format for C code (!) * Makes the plugin syntax and semantics clearer * Requires some XML tools ## Example Plugin: Valve * Simulation of a tube amplifier * Uses limited history to do some filtering as well as distortion * Based on a Norwegian [thesis]( * How does it sound? ## Example: Compression/Expansion * Idea: Compression — try to hold output volume level more steady as function of input volume * Idea: Expansion — try to make changes in input volume more pronounced on output * Typical implementation: two different linear (in log space because volume) gain functions with a "knee" * Effect at knee must not be instantaneous, else signal will distort: "attack" time to move to upper curve, "decay" time to move back to lower curve * Used a lot because "professional sound" and because always-high volume is desirable * Demo: Audacity built-in
Last modified: Tuesday, 30 April 2019, 4:03 PM